Guidelines for Designing AI Technologies to Support Adult Learning

Synthesis: Drawing on longitudinal deployment data from the National AI Institute for Adult Learning and Online Education (AI-ALOE), this DIS 2026 paper synthesizes 19 empirically grounded design guidelines for AI-powered adult learning technologies. The guidelines span cognitive, social, and teaching presence dimensions and are derived from reflexive thematic analysis of ~1,600 stakeholder statements across seven deployed systems. The work also provides a heuristic evaluation method and an interactive guideline exploration tool.

Context and Motivation

AI-powered educational technologies have demonstrated benefits but are predominantly designed and evaluated in K-12 contexts. Adult learners differ motivationally and contextually: they are often self-directed, goal-oriented (career advancement, reskilling), and must balance education with employment and family responsibilities. Existing systems inherit design patterns from K-12 that are poorly aligned with adult learning constraints.

This paper examines seven AI-powered technologies deployed within AI-ALOE, a US National AI Institute:

• Apprentice Tutors — web-based ITS for adult math/STEM practice
• iTELL — Intelligent Texts for Enhanced Lifelong Learning
• Ivy — interactive video-based AI coaching for procedural skills
• Jill Watson — RAG-powered Q&A agent for online courses
• SAMI — social agent for online discussion forums
• SMART — Student Mental Model Analyzer for Research and Teaching
• VERA — conceptual modeling tool for guided inquiry

Methodology

Using reflexive thematic analysis, the team analyzed approximately 1,600 stakeholder statements from focus groups with learners and instructors, technical artifacts, and progress reports. Statements were organized through affinity diagramming into sub-themes, needs statements, and finally 19 design guidelines framed as "AI tools should..." Each guideline is labeled by Community of Inquiry dimension (cognitive/social/teaching presence) and most-impacted stakeholder.

The 19 Design Guidelines

The guidelines span four categories:

Cognitive Presence

• G2: AI tools should be accessible and fit into the busy lives of adult learners (mobile-friendly, offline-capable, affordable)
• G3: AI tools should be informed by learning science and learning theories
• G4: AI tools should be easy to understand and frictionless to use

Teaching Presence

• G7: AI tools should support learner Motivation and engagement
• G8: AI tools should align with established instructional best practices
• G9: AI tools should support diverse pedagogical strategies
• G11: AI tools should personalize the learning experience
• G13: AI tools should provide substantive educational support (not just surface-level assistance)
• G16: AI tools should align with instructors' personal instructional approach
• G17: AI tools should provide meaningful Feedback and explanations

Social Presence

• G18: AI tools should scaffold and support learners in developing their social competencies
• G19: AI tools should foster social connection and community

Cross-Cutting

• G1: AI tools should be transparent about data practices (collection, storage, access)
• G6: AI tools should provide scaffolded support that adapts as learners progress
• G12: AI tools should connect content to real-world problems that matter to adult learners
• G14: AI tools should support learning engineering through actionable data
• G15: AI tools should integrate easily with existing educational ecosystems

Key Findings

• Stakeholder priorities diverge: Instructors focused on usability (G4) and instructional alignment (G16); students emphasized educational support (G13) and community-building (G18, G19); researchers prioritized learning theories (G3) and best practices (G8-G10).
• No single technology satisfied all 19 guidelines — but the broader AI-ALOE ecosystem collectively covered the full set.
• Personalization (G11) had low satisfaction across systems: most deployed surface-level personalization (adapting examples, knowledge checks) rather than deeper adaptations like task sequencing or difficulty calibration.
• Data transparency (G1) and social/community features (G18, G19) consistently scored lowest across the heuristic evaluation.

What this means for practice

• Instructors. Run the paper's 19 guidelines as a heuristic evaluation before adopting any adult-learning AI tool, and expect no single tool to satisfy all 19 — the AI-ALOE ecosystem covered the full set only collectively.
• Designers. Treat adult-life constraints as prerequisites rather than features: mobile access, offline capability, and affordability came directly from learners balancing work, family, and study.
• Designers. Ground tools in andragogy, not just Pedagogies and Teaching Strategies: adult learners are self-directed and problem-oriented, and tools that inherit K-12 design patterns fit their constraints poorly.
• Designers. Fix the lowest-scoring areas first: data transparency and social/community features scored lowest across the deployed systems, and personalization stayed surface-level — adapting examples and knowledge checks rather than task sequencing or difficulty calibration.

Limitations

• The corpus comes from a single institute: 15 research-team presentations, transcripts from 17 focus groups, and 3 sets of cross-team feedback (about 1,600 statements) covering seven AI-ALOE technologies, so the guidelines reflect that ecosystem's deployments rather than adult learning generally.
• The synthesis is interpretive by design — the authors state that another research team drawing on different experiences and values might arrive at an alternate set of themes or guidelines.
• The heuristic evaluation tested systems against the guidelines rather than measuring learning; no guideline is validated against learner performance data, and the guideline exploration tool is presented as a demonstration of utility.
• Guideline-level analysis counts statements rather than effects: grounding in learning science (G3) was raised entirely by research teams, and instructors and learners never explicitly discussed the value of theoretical or empirical evidence, so the set mixes concerns no learner voiced with those they prioritized most.

Citation

Reddig, J., Smith, G. R., Jr., Ahmadzadeh Siyahrood, S., Morris, W. G., Bae, Y., Crutcher, K., et al. (2026). Guidelines for Designing AI Technologies to Support Adult Learning.